Table of Contents
Prologue
Laravel’s AI development tools and ecosystem have evolved significantly in 2026. The framework now provides a first-party SDK for AI development, and several community packages have emerged, addressing particular issues the framework does not fully resolve. The critical question surrounding these packages is not which package is better but which architectural problem they aim to solve. A provider abstraction library, an agent framework, and a protocol-level utility like MCP are not mutually exclusive but often solve divergent problems and can be used together.
This article was last updated on October 5th, 2026.
4 AI packages for Laravel – this is a decision guide, not a ranking.
1. 🧩 Laravel AI SDK, the First-Party Starting Point
Overview
Laravel released the first beta of its AI SDK in February 2026 and released v1.0 on September 23, 2026, marking its first stable release.
composer require laravel/ai
What it covers (based on the documentation as of this writing): unified API across multiple AI providers, agents, structured output, streaming responses, tool/function calling, conversation storage with a database-backed store and migrations provided by the SDK, embeddings and a SimilaritySearch tool for vector search, and image, audio, and transcription generation. Version 1.0 specifically added classification (for use cases that aren’t really “writing” like routing or tagging) and tool approvals (a human-in-the-loop sign-off mechanism before a tool executes) and implements the AG-UI protocol used by clients like CopilotKit.
Where It Fits
This should be your starting point for general AI application development in Laravel. It’s first-party, actively developed, and deeply integrated with the framework (migrations, config, facades).
What To Watch
v1.0 contained breaking changes from the beta, particularly in conversation storage and agent middleware, so if you adopted it early, you’ll need to budget real time for the upgrade guide, not a quick one composer update.
2. 🔌 Prism, Provider Abstraction Done Well
Overview
composer require prism-php/prism
Prism is an actively maintained independent package (written by TJ Miller, 2,200+ GitHub stars, requires PHP 8.2+ and Laravel 11+), focused exclusively on delivering a consistent, fluent API for working with laravel/laravel’s default set of providers, text generation, multi-turn conversations, and tool calling, without having to roll your own implementations for each provider.
Where It Fits
If you need a lighter-weight provider abstraction layer that doesn’t tie you to laravel/laravel’s first-party AI stack, or if you’re working on Laravel 9/10 and the required PHP version for laravel/laravel’s official SDK doesn’t work for you, then this is a great option. It’s an actively developed, independent library with a growing following; other packages leverage it as a foundation ( LarAgent is built using it, for example), and more tools are being built on top of it ( Atlas is an orchestration layer built directly on top of it).
What To Watch
Prism is an independent package not affiliated with the Laravel core team. That’s not a criticism in any way; it has active development and community support, but you should be aware that it can’t receive first-party support
3. 🤖 LarAgent, When You Need Agent-First Architecture
Overview
composer require maestroerror/laragent
LarAgent (Laravel 10.x+ and PHP 8.3+) has a fundamentally different approach than the tools discussed so far. It’s an Eloquent-style syntax for declaring and managing AI agents, not general-purpose AI calls. You define an agent class in the same way you would a model, register tools as class methods, receive built-in chat history management, and have native Laravel event hooks for the agent lifecycle. It supports OpenAI, Anthropic, Groq, Gemini, and Ollama, and can use Prism underneath for additional provider coverage.
Where It Fits
Applications that are fundamentally agent-shaped: a support bot that needs persistent memory and tool access, a multi-step autonomous workflow, something that needs to behave like a long-running agent rather than respond to one-off prompts.
What To Watch
This is a narrower, more opinionated tool than the AI SDK or Prism. If your actual need is “call an LLM and get structured output back,” LarAgent’s agent-first design is more machinery than you need.
4. 🔗 Laravel MCP, Interoperability, Not Another LLM Wrapper
Overview
composer require laravel/mcp
One of the easiest to confuse with the others: the Laravel MCP is not a tool for calling AI models. Instead, it is the official Laravel package for building MCP servers exposing your business logic as tools – callable from external AI clients (Claude, Cursor, or any other MCP-compatible agent). The package implements Laravel’s routing conventions (Mcp::local(), Mcp::web()), provides Artisan commands to scaffold servers, tools, resources, and prompts, and includes a first-party test client for Pest/PHPUnit.
Where It Fits
Expose your Laravel app’s capabilities to AI agents – to let an external assistant query your order system, trigger a workflow, or read application state via a standardized protocol, instead of having AI running inside your app.
What To Watch
Do not confuse it with the AI SDK, Prism, or LarAgent – Laravel’s first-party package for building MCP servers to be used by external agents.
Decision Matrix
| Tool | Best Fit | Problem Solved | Choose This When |
|---|---|---|---|
| Laravel AI SDK | General AI application development | Unified provider API, agents, structured output, streaming, conversations | You want the first-party, actively-developed default |
| Prism | Lightweight provider abstraction | One interface across multiple LLM providers | You need provider flexibility without the full AI SDK stack, or you’re on an older Laravel version |
| LarAgent | Agent-centric applications | Persistent, stateful AI agents with memory and tools | Your feature is fundamentally an agent, not a single prompt/response |
| Laravel MCP | Exposing your app to AI clients | Lets external AI agents call your app’s tools | You’re building the server side of an AI integration, not the client side |
Before You Choose
A few things to consider regardless of which tool you decide to go with:
- Provider lock-in: How difficult would it be to switch providers if pricing, performance, or quality changes?
- Laravel-version compatibility: Does the package require PHP 8.3+ and Laravel 10+ (LarAgent), or is it more flexible (Prism’s requirements are less strict)? Does it support your actual stack or just the bleeding edge?
- Structured output and tool calling: Does the package provide what you need for your specific feature, or do you have to roll your own on top of text generation?
- Streaming: Does your frontend need per-token streaming, and does the package support it?
- Testing and observability: Can you mock agent/tool responses in your test suite, and can you observe what the model did in production if something goes wrong?
- Cost and failure handling: What is the fallback when a provider call fails or times out, and how will you control unexpected costs?
Conclusion
There’s no single “best” Laravel AI package because each one is designed for different use cases. As such, it is better to describe a practical scenario in which one might be chosen over another. As of right now, I would personally use the Laravel AI SDK as my primary tool for building new AI features due to it being developed officially and addressing the most common use cases. If I needed more lightweight abstractions over specific providers, I would consider using Prism, especially when I need its provider-abstraction approach and its current requirements fit the application. In cases where I need to build an agent with some persistent memory, I would consider using the LarAgent. Finally, if I needed to expose my application to AI clients rather than consume it, I would go with Laravel MCP, as it serves an entirely different purpose compared to the other four packages mentioned above.
Related Reading
Official Sources
- Announcing Laravel Boost
- Laravel AI SDK v1.0 announcement (Sept 23, 2026)
- Laravel AI SDK documentation
- Prism PHP repository
- LarAgent repository
- Laravel MCP package
